public expenditure and economic growth: a disaggregated analysis for developing countries*

PUBLIC EXPENDITURE AND ECONOMIC GROWTH:A DISAGGREGATED ANALYSIS FOR

DEVELOPING COUNTRIES*

byNILOY BOSE

University of Manchester and University of Wisconsin-MilwaukeeM. EMRANUL HAQUE

andDENISE R. OSBORN†

University of Manchester

In this paper, we examine the growth effects of government expenditurefor a panel of 30 developing countries over the 1970s and 1980s, with aparticular focus on disaggregated government expenditures. Our meth-odology improves on previous research on this topic by explicitly recog-nizing the role of the government budget constraint and the possiblebiases arising from omitted variables. Our primary results are twofold.First, the share of government capital expenditure in GDP is positivelyand significantly correlated with economic growth, but current expendi-ture is insignificant. Second, at the disaggregated level, governmentinvestment in education and total expenditures in education are the onlyoutlays that are significantly associated with growth once the budgetconstraint and omitted variables are taken into consideration.

1 Introduction

The recent revival of interest in growth theory has also revived interest amongresearchers in verifying and understanding the linkages between fiscal policiesand economic growth. Over the past decade and a half, a substantial volumeof empirical research has been directed towards identifying the elements ofpublic expenditure (at its aggregate and disaggregate levels) that bear signifi-cant association with economic growth. This empirical literature varies interms of data sets and econometric techniques, and often produces conflictingresults.1 Explanations offered to account for these varied and conflictingresults can broadly be divided into two categories. According to the first, it is

* Manuscript received 25.5.06; final version received 29.5.07.†The third author gratefully acknowledges financial assistance from the Economic and Social

Research Council (UK) under grant number L138251030. The second author gratefullyacknowledges financial assistance from the Commonwealth Scholarship Commission. Theauthors would like to thank Jonathan Temple for his insightful comments in relation to thisresearch and the referees for their constructive suggestions. The permission to use WorldBank Archives at Washington, DC, for data collection is also gratefully acknowledged.

1Consider, for example, the association between government size (as measured either by the levelof total public expenditure or by the level of public consumption expenditure) and eco-nomic growth. According to some studies, such association is significant and positive(Ram, 1986; Romer, 1989, 1990). The same association has been found to be significant and

The Manchester School Vol 75 No. 5 September 20071463–6786 533–556

© 2007 The AuthorsJournal compilation © 2007 Blackwell Publishing Ltd and The University of ManchesterPublished by Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK, and 350 Main Street, Malden, MA 02148, USA.

533

the differences in the set of conditioning variables across studies that areresponsible for the lack of consensus in the results (Levine and Renelt, 1992).In contrast, the second category consists of a handful of studies (Helms, 1985;Mofidi and Stone, 1990; Kneller et al., 1999) that suggest this variation in theresults, in part at least, reflects the widespread tendency among researchers toignore the implications of the government budget constraint for their regres-sions. In particular, the latter view emphasizes the need to consider both thesources and the uses of funds simultaneously for a meaningful evaluation ofthe effects of taxes or expenditures on economic growth.

In addition to producing conflicting views, the existing literature dis-plays a disturbing trend. Most of the conclusions drawn regarding the growtheffects of public spending are based either on the experiences of a set ofdeveloped countries or on the basis of large samples consisting of a mixtureof developed and developing countries. Accordingly, there remains little byway of understanding the process by which public expenditure policies shapethe growth prospect for developing countries.2 This trend has continueddespite the long-standing view among development experts not only thatthere exists a significant difference in the composition of public expenditurebetween developed and developing countries, but also that the difference isprofound in the way in which public expenditures shape the outcome in thesetwo sets of countries.3

The primary objective of this paper is to examine the growth effects ofpublic expenditure by sector for a set of developing countries, paying attentionto the ‘sensitivity’ issue arising from initial conditions and conditioning vari-ables while also recognizing the possible existence of correlation between theexpenditures in different sectors that may result in spurious coefficients in thegrowth equation due to omitted variables. Here, we are not interested inthe financing of any particular public expenditure per se, but we include theimportant financing variables (government budget surplus/deficit and taxrevenue) to avoid the coefficient biases that would result from their omission(Kneller et al., 1999). Further, where government expenditure components arefound to be individually significant, we include them jointly to investigatewhether their apparent individual roles are genuine, or spurious in the sense ofbeing attributable to other components with which they are correlated. Inother words, from an econometric perspective, we again control for possible

negative in other studies (e.g. Landau, 1983, 1985, 1986; Grier and Tullock, 1989;Alexander, 1990; Barro, 1990, 1991). Yet other studies have found this association tobe insignificant or fragile (e.g. Kormendi and Meguire, 1985; Levine and Renelt, 1992).A similar variation in results can also be observed among studies which look for the growtheffects of public expenditures at disaggregated levels.

2Notable exceptions include the contributions by Landau (1986), Devarajan et al. (1996) andMiller and Russek (1997).

3For example, empirical evidence suggests that low- and high-income countries differ signifi-cantly in terms of effective use of infrastructure resources and this efficiency effect mayaccount for nearly 40 per cent of the growth differential between these two sets of countries(Hulten, 1996). Also, please refer to the World Bank Report (1988) for further details.

The Manchester School534

© 2007 The AuthorsJournal compilation © 2007 Blackwell Publishing Ltd and The University of Manchester 2007

omitted variable bias that will result should any component of governmentexpenditure that is important for growth be excluded from the model. Thus, onone hand, by focusing our attention exclusively on developing countries and,on the other, by recognizing the full implication of the government budgetconstraint and the potential collinearity between the expenditure components,we aim to make a distinct and meaningful contribution to the literature.4

Our disaggregated analysis is also valuable from the policy perspective.Our results for the growth effects of public expenditures by individual sectorsof the economy give rise to information that is particularly useful for devel-oping countries, which are resource constrained and where the allocation oflimited public resources between sectors is an issue of paramount importance.In this regard, our main contribution is the finding that education is the keysector to which public expenditure should be directed in order to promoteeconomic growth. This result is novel and overturns previous findings ofnegative or insignificant positive effects of education expenditure on growthfor developing countries (Landau, 1986; Devarajan et al., 1996; Miller andRussek, 1997). However, as argued above, our analysis is more satisfactoryfrom an econometric perspective than earlier studies.

Our two principal empirical findings can be summarized as follows.

1. The share of government capital expenditure in GDP is positively andsignificantly correlated with economic growth, while the growth effect ofcurrent expenditure is insignificant for our group of countries.

2. At the disaggregated level, government investment in education and totalexpenditures in education are the only outlays that remain significantlyassociated with growth throughout the analysis.

Other findings of our analysis are the following.

3. Although public investments and expenditures in other sectors (transportand communication, defence) initially have significant associations withgrowth, these do not survive when we incorporate the governmentbudget constraint and other expenditure components into the analysis.This is in contrast to the prevailing view (e.g. Easterly and Rebelo, 1993)that the correlation between transport and communication expenditureand growth is one of the most robust findings.

4. The private investment share of GDP is associated with economic growthin a significant and positive manner.

5. There is strong evidence that a government budget deficit gives rise toadverse growth effects.

4The primary objective of our study overlaps with those by Landau (1986) and Devarajan et al.(1996). These studies, however, only focus on the expenditure side of the budget constraintand ignore the revenue side. The study by Miller and Russek (1997) pays relatively moreattention to the government budget constraint. In contrast to ours, the focus of their studyhas been to investigate how the growth effect of a public expenditure varies with the modeof financing such expenditure.

Public Expenditure and Economic Growth 535


The remainder of the paper is organized as follows. Section 2 discussesour data and its sources. Section 3 presents a baseline analysis of the impactof government expenditure categories on growth, which is extended inSection 4 to examine the implications of omitted variable bias and thegovernment budget constraint. Section 5 concludes.

2 Data and Variables

We use data on public expenditures for both current and capital expendi-tures,5 at aggregated and disaggregated levels, for 30 developing countries6

using decade averages over the period 1970–90. Our data are drawn from theCentral Government Consolidated accounts for these countries.

Although the Government Financial Statistics, published annually by theInternational Monetary Fund (IMF), is a primary source for data on gov-ernment expenditures at aggregate level, it is not appropriate for our study.This is because, although Government Financial Statistics publishes data fortotal expenditures by sector and aggregate capital and current expenditures,it does not provide any data for sectoral current and capital expenditures.Consequently, the usefulness of this data source is limited for our analysisas, in addition to the aggregate capital and current expenditures, wewish to study the effects of capital and current expenditures by sector (e.g.defence, education, health, agriculture, transport and communication, andmanufacturing).

To overcome this problem, we constructed a data set after consulting alarge collection of World Bank Country Economic Reports and Public Expen-diture Reviews.7 From these, information about the central government’stotal, current and capital expenditures by sector was extracted over 1970–90for 30 developing countries, and hence these countries constitute our sample.Although we aimed at collecting the data for as many countries and as manyyears as possible, we were constrained by unavailability of the relevant docu-ments in the World Bank Archives for many countries after 1990. Further,due to the number of missing observations for the government expenditures

5We have followed the Government Financial Statistics Yearbook (published by IMF, 1986)guidelines for classifying expenditures into current and capital expenditures.

6The countries are listed in the Appendix. According to the World Bank classification, the samplehas 17 low-income countries, nine lower middle-income countries and three upper middle-income countries. Only the Bahamas falls slightly above middle-income countries.

7In an earlier exercise, Easterly and Rebelo (1993) collected data on public investment by sectors.We differ from this existing data set on two grounds. First, our data set includes informa-tion on both public investment expenditures and current expenditures by sector. Second,the measure of public investment used by Easterly and Rebelo also includes investment bypublic enterprises. In contrast, we strictly follow the Government Financial Statistics guide-lines and exclude public enterprise investments. We acknowledge that this narrower defi-nition may give rise to some bias in the results. At the same time (as acknowledged by theauthors themselves) the measure used by Easterly and Rebelo (1993) creates a tendency tooverstate public investment by including investments by public firms that have activitiesand goals similar to those of the private sector.



data, it is impractical to conduct an analysis at the annual frequency andhence our analysis is conducted using decade averages.

Data for other variables have been drawn from two different datasources. Initial GDP per capita, population, initial human capital, life expect-ancy, political instability, private investment, initial trade ratio, black marketpremium and the terms of trade have been extracted from the Barro and Lee(1994) data set. Growth of GDP per capita, agriculture’s share in GDP andbroad money (M2) have been extracted from the World Development Indi-cators of the World Bank.

Unless we state otherwise, a data point for a variable corresponds to thedecade average value (1970–79, 1980–89) of that variable. The details of thevariables and their data sources are included in the Appendix.

3 Baseline Results

To start with, we classify the variables into three distinct sets: I, M and Z. Theset I consists of variables that commonly appear as conditioning variablesin growth regressions. The set Z includes variables that often have beenincluded in previous studies as indicators for monetary policies, trade policiesand market distortion. Finally, the set M consists of variables that are ofparticular interest for the present study, namely central government expen-ditures and their major components at aggregate and sectoral levels. Thesevariables are expressed as percentages of GDP. In total, we consider 20 suchvariables, as detailed in the Appendix. To make our tables digestible,however, we do not report results for variables with no significant associationwith growth at the most elementary stage of our analysis, i.e. in the baseregression (1) below.

Operationally, we use a panel set-up in which the dependent variable(growth rate in real GDP per capita, GRit) is observed twice (as decadeaverages) for each country for 1970–79 and 1980–89. The system includes aseparate constant term, b0t, for each decade. The other coefficients are con-strained to be the same for both time periods. Panel estimation is carried outby the seemingly unrelated regression method, with two equations for eachcountry (one equation for each decade). Thus, the disturbance term, uit, forcountry i at time t, is allowed to be correlated with term uit′ for the samecountry at the different date, t′. The variance of uit varies with t but not withi. In practice, the estimated correlations of the error terms across the timeperiods turn out to be small and insignificant (see Tables 1–3).

3.1 Base Regressions

Initially, we examine whether the variables of interest (i.e. the elements of theset M) are significantly correlated with growth after controlling for the Ivariables. For this, we run a series of base regressions each of which includesall conditioning (I) variables and one government expenditure (M) variable:



GR ,it t jI

j itM

it itj

I M u= + + +=

∑β β β01

6

(1)

Following Levine and Renelt (1992), Barro (1991) and Barro and Sala-i-Martin (1995, 1999), we include log of initial GDP per capita, initial schoolenrolment ratio,8 private investment share of GDP, log of life expectancy andan index of political stability in the set I.9 It has been emphasized by a numberof studies (e.g. Cashin, 1995; Kocherlakota and Yi, 1997) that, while theprovision of public goods is growth-enhancing, the distortionary taxes thatneed to be raised to fund the provision of the same public goods may havegrowth-diminishing effects. Accordingly, it is necessary to control for taxrevenue in order to make a proper assessment about the growth effects ofpublic spending. Keeping this view and the primary objective of this paper inmind, we have also included tax revenue as a percentage of GDP in the set I.Accordingly, the set I of the base regression (1) embodies a central idea of thenew growth literature, namely that human capital and institutional factorsare important determinants of economic growth. In addition, through inclu-sion of initial GDP, the above model also controls for possible effects ofconvergence on output growth.

Table 1 summarizes the results from the base regression (1). Out of the20 categories of public expenditure examined, we report the results only forthe six categories (total investment, investment in education, investment intransport and communication, total expenditure on education, total expen-diture on transport and communication, and total expenditure on defence)that we find to display a significant association with growth, using a 10 percent significance level.

We open the discussion with our results for the I variables. Among thisset, only private investment demonstrates a significant association withgrowth. This is in congruence with the basic prediction of neoclassical growththeory, and is supported by a number of previous empirical studies (e.g.DeLong and Summers, 1991; Levine and Renelt, 1992; Mankiw et al., 1992).Some other results, however, are less in tune with the theoretical predictions.For example, our analysis shows no sign of convergence among this group ofcountries. We suspect this may be due to the fact that our sample includesa number of poor countries (such as Sub-Saharan countries), which experi-enced dismal growth performances (often negative growth rates) over a

8We also considered average schooling years as a proxy for the human capital stock. However,we dropped this variable from our analysis due to the absence of data for a quarter of thecountries in our sample.

9Levine and Renelt (1992) also include average annual population growth rate in the set I, but wedropped it from the analysis since it was always insignificant, perhaps due to the lack ofvariability in its values. We did, however, verify that all our results remain unaltered whenthis variable is included in the analysis.



Ta

ble

1G

ro

wth

Reg

res

sio

ns

wit

hC

entr

al

Go

ver

nm

ent

Ex

pen

dit

ur

es

Cap

ital

expe

ndit

ure

Edu

cati

onin

vest

men

t

Tra

nspo

rtan

dco

mm

unic

atio

nin

vest

men

tE

duca

tion

expe

ndit

ure

Tra

nspo

rtan

dco

mm

unic

atio

nex

pend

itur

eD

efen

ceex

pend

itur

e

Gov

ernm

ent

expe

ndit

ure

0.17

1***

(0.0

56)

1.51

6***

(0.4

31)

0.38

9*(0

.206

)0.

681*

**(0

.239

)0.

394*

*(0

.191

)0.

257*

(0.1

35)

Iva

riab

les

Tax

reve

nue

-0.0

30(0

.055

)0.

041

(0.0

52)

-0.0

20(0

.062

)-0

.096

(0.0

68)

-0.0

44(0

.067

)-0

.003

(0.0

64)

Pri

vate

inve

stm

ent

0.26

5***

(0.0

53)

0.24

2***

(0.0

53)

0.24

6***

(0.0

59)

0.28

3***

(0.0

55)

0.24

9***

(0.0

58)

0.29

4***

(0.0

58)

Init

ialG

DP

per

capi

ta0.

005

(0.0

03)

0.00

4(0

.003

)0.

003

(0.0

03)

0.00

5(0

.003

)0.

003

(0.0

03)

0.00

4(0

.003

)In

itia

lhum

anca

pita

l-0

.012

*(0

.007

)-0

.011

*(0

.006

)-0

.014

**(0

.007

)-0

.009

(0.0

07)

-0.0

16**

(0.0

07)

-0.0

13*

(0.0

07)

Init

iall

ife

expe

ctan

cy0.

076

(0.0

70)

0.05

0(0

.068

)0.

116

(0.0

78)

0.01

5(0

.075

)0.

136

(0.0

79)

0.09

3(0

.070

)P

olit

ical

inst

abili

ty-0

.007

(0.0

20)

-0.0

06(0

.020

)-0

.014

(0.0

20)

-0.0

04(0

.020

)-0

.016

(0.0

20)

-0.0

25(0

.019

)R

20.

500.

510.

440.

510.

460.

560.

520.

570.

530.

560.

550.

64O

bser

vati

ons

3029

2928

2825

3029

2928

2825

Reg

ress

ion

test

(pva

lue)

61.6

02(0

.000

)67

.891

(0.0

00)

46.4

65(0

.000

)56

.926

(0.0

00)

46.4

65(0

.000

)65

.499

(0.0

00)

AR

(1)

coef

ficie

nt(p

valu

e)0.

018

(0.9

22)

0.00

2(0

.993

)0.

176

(0.3

42)

0.13

9(0

.462

)0.

162

(0.3

92)

0.16

4(0

.413

)

Not

es:

The

colu

mn

head

ing

show

sth

esp

ecifi

cgo

vern

men

texp

endi

ture

cate

gory

(Mit)u

sed

inth

ere

gres

sion

.Est

imat

ion

isby

the

seem

ingl

yun

rela

ted

regr

essi

onte

chni

que,

whi

chal

low

sth

eer

ror

term

tobe

corr

elat

edac

ross

the

two

deca

des

and

toha

vea

diff

eren

tva

rian

cein

each

peri

od.

The

depe

nden

tva

riab

leis

grow

thra

tein

real

GD

Ppe

rca

pita

.St

anda

rder

rors

ofco

effic

ient

sar

esh

own

inpa

rent

hese

s.T

hefir

stR

2is

for

the

1970

san

dth

ese

cond

R2

repo

rted

isfo

rth

e19

80s.

Sim

ilarl

yth

efir

stnu

mbe

rof

obse

rvat

ions

isfo

rth

e19

70s

and

the

seco

ndnu

mbe

rof

obse

rvat

ions

repo

rted

isfo

rth

e19

80s.

The

num

ber

ofob

serv

atio

nsdi

ffer

sac

ross

mod

els

due

toth

ela

ckof

avai

labi

lity

ofso

me

expl

anat

ory

vari

able

sfo

rsp

ecifi

cco

untr

ies.

The

seri

alco

rrel

atio

nco

effic

ient

isth

eA

R(1

)va

lue

ina

regr

essi

onof

resi

dual

sfo

r19

70s

and

thos

efo

r19

80s,

wit

hth

ep

valu

ebe

ing

that

for

the

Bre

usch

–Pag

ante

st,w

here

the

null

hypo

thes

isis

that

the

resi

dual

sof

the

equa

tion

sfo

rtw

ode

cade

sar

eun

corr

elat

ed.T

here

gres

sion

test

isa

Wal

dc2

test

wit

hth

enu

llhy

poth

esis

that

the

coef

ficie

nts

are

zero

.For

the

coef

ficie

nts,

*ind

icat

essi

gnifi

cant

at10

per

cent

,**s

igni

fican

tat

5pe

rce

ntan

d**

*sig

nific

ant

at1

per

cent

.



prolonged period of time.10 Surprisingly, initial human capital is found tohave a negative effect on growth, with this sometimes being significant.11 Interms of direction, the relationships between growth and the remaining twoconditioning variables accord well with theoretical predictions, but neither ofthese associations is significant for this group of countries.

As already noted, our preliminary analysis indicates that the GDP sharesof only six out of 20 categories of public spending individually display anassociation with economic growth. Table 1 shows the levels of significanceacross these to be varied.12 The most significant associations are obtained fortotal capital expenditure, for total expenditure in the education sector and forinvestment expenditure in the education sector. The significant associationbetween the share of central government capital expenditure in GDP andeconomic growth is not entirely surprising in the light of the conclusionsdrawn by previous studies (e.g. Easterly and Rebelo, 1993; Cashin, 1995; dela Fuente, 1997) that are based on either developed countries or a large poolof developed and developing countries. However, to our knowledge, Landau(1986) is the only panel study that included total capital expenditure in theregression for developing countries, but found its association with growth tobe insignificant. Thus, our result here contains new information.

Our result on total education expenditure differs from conclusionsdrawn by previous studies, irrespective of whether these are based on data fora large pool of countries (e.g. Barro and Sala-i-Martin, 1995, 1999) or devel-oping countries (e.g. Landau, 1986; Devarajan et al., 1996). These earlierresults indicate that the association of this variable with growth is eitherinsignificant or non-robust.

Our result regarding the association between investment expenditure inthe education sector and economic growth also merits some comment. Due tothe lack of readily available data, the analysis of the impact of this variable ongrowth is almost non-existent in the literature. To our knowledge, the onlyexception is Easterly and Rebelo (1993), who study a large pool of developedand developing countries. Not only do we find investment in education to behighly significant, but the magnitude of the effect of this variable on growthis considerable: a 1 percentage point increase in central government invest-ment in education in relation to GDP is associated with an increase in theaverage annual growth rate of real GDP per capita by 1.5 percentage points.Although not significant in their case, Easterly and Rebelo (1993) find simi-

10In the growth literature (e.g. Azariadis and Drazen, 1990) often these countries have beenreferred to as the countries in a ‘development trap’.

11This result draws support from a number of recent cross-country studies that found economicgrowth to be uncorrelated with increases in educational attainment (see, for example,Benhabib and Spiegel, 1994; Gemmell, 1996; Temple, 1999a, 1999b; Pritchett, 2001).

12The fact that all six expenditure variables reported in Table 1 appear significant is somewhatstriking. This is not always the case in previous studies and such difference in result isindicative of the real value to the data that we have collected over the commonly used IMFsources.



larly large effects for investment in education. The explanation for this effectmay lie in the strong externalities of investment in education in raising theproductivity of both human and physical capital. Theoretical justification ofthis view is readily available in the new growth literature.

Results for the other three expenditure variables draw mixed supportfrom the existing literature. For example, the positive and significant asso-ciation between total expenditure in the transport and communication sectorand growth finds support in the study by Aschauer (1989). Support for thepositive association between investment expenditure in the transport andcommunication sector and growth can be obtained in the study by Easterlyand Rebelo (1993). We, however, find this association to be significant only atthe 10 per cent level. Finally, our preliminary analysis suggests a positive andsignificant (at the 10 per cent level) association between defence spending andgrowth. In the existing literature, this association has sometimes beenreported as positive and significant (Benoit, 1978; Fredriksen and Looney,1982). At the same time, other studies have found it to be negative (Deger andSmith, 1983; Knight et al., 1996), while in yet other studies the growth effectof defence expenditure has been found to be neutral (Biswas and Ram, 1986).

3.2 Robustness Checks

The robustness of the results from the base regression (1) to the inclusion ofZ variables is now examined, focusing only on the M variables that areassociated with growth in a significant manner and included in Table 1. Thisanalysis is conducted in two stages. First, following Easterly and Rebelo(1993), we expand the set of regressors to include the ratio of broad money(M2) to GDP in 1970 and the trade share13 of GDP in 1970 (TR):

GR M TR,it t jI

j itM

itZ

itZ

it itj

I M u= + + + + +=

∑β β β β β0 1 21

6

2 (2)

The purpose of including these variables is to control for the effects ofmonetary policy and the degree of openness which, according to previousstudies (e.g. Levine and Renelt, 1992; King and Levine, 1993), are potentiallyimportant correlates of economic growth. Next, we expand the set of regres-sors to include other variables:

GR M TR BMP TT,it t jI

j itM

itZ

itZ

itZ

itZ

itI M u= + + + + + + +β β β β β β β0 1 2 3 42 iitj =∑

1

6

(3)

More specifically, we include the black market premium (BMP) and thegrowth rate of the terms of trade (TT) in (3). These control for marketdistortions and capture the adverse effect of external shocks that a number of

13Rodrik (1998) has also argued in favour of including the trade share when estimating therelationship between fiscal variables and growth.



countries in our sample experienced during the period of our analysis. Thesetwo variables have also appeared as significant correlates of growth in pre-vious studies (e.g. Fischer, 1993; Devarajan et al., 1996; Barro and Sala-i-Martin, 1999). The results are reported in Table 2.

In the spirit of Levine and Renelt (1992), we certify that the variable underconsideration has a robust association with economic growth if the coefficientof the M variable remains significant and of the same sign as in Table 1 afterinclusion of these additional variables. As our results indicate, none of the sixexpenditure variables fails the robustness test. In fact, in most cases, weobserve an improvement in the level of significance. In contrast, for thecountries in our sample, of the four Z variables only the growth of the terms oftrade shows significant association with economic growth in some cases.

Therefore, the results of the base regression in Table 1 have not beenunduly distorted by omission of variables capturing monetary policies, tradepolicies or market distortions.

4 Omitted Variables and the Government Budget Constraint

4.1 The Government Budget Constraint

We noted in Section 1 that almost all previous studies of the associationbetween government expenditure and growth are subject to potential biasesbecause they omit variables that enter the government’s budget constraint.This is the case also for the regressions (1)–(3) above, whose results have beensummarized in Tables 1 and 2.

Kneller et al. (1999) discuss the importance of the government budgetconstraint in the context of the growth effects of fiscal policy for developedcountries. Our discussion primarily follows Kneller et al. (1999).14 General-izing the notation of Section 3, let Mj,it be the fiscal variable j relating tocountry i at time t. The set of fiscal variables include government expenditure,government revenue and budget deficit items. If there are m such distinctelements, then the government budget constraint implies the identity

Mj itj

m

, ==

∑ 01

Allowing each element to have an impact on growth leads to a generalizationof the growth regression (1) as

GR , ,it t jI

j it jM

j it itj

m

j

I M u= + + +==

∑∑β β β011

5

(4)

14Miller and Russek (1997) make arguments similar to those of Kneller et al. (1999), but they donot consider omission bias in their econometric analysis (see footnote 4).



Ta

ble

2R

ob

ust

nes

sC

hec

ks

for

Eff

ects

of

Go

ver

nm

ent

Ex

pen

dit

ur

e

Tot

alin

vest

men

tE

duca

tion

inve

stm

ent

Tra

nspo

rtan

dco

mm

unic

atio

nin

vest

men

tE

duca

tion

expe

ndit

ure

Tra

nspo

rtan

dco

mm

unic

atio

nex

pend

itur

eD

efen

ceex

pend

itur

e

Gov

ernm

ent

expe

ndit

ure

0.20

1***

(0.0

57)

0.20

1***

(0.0

60)

1.80

3***

(0.4

46)

1.99

7***

(0.4

46)

0.43

1**

(0.2

13)

0.47

1**

(0.2

20)

0.67

9***

(0.2

45)

0.67

5***

(0.2

48)

0.47

1**

(0.1

98)

0.51

5**

(0.2

05)

0.32

4**

(0.1

49)

0.37

2**

(0.1

49)

Zva

riab

les

Init

ialM

20.

024

(0.0

26)

0.01

7(0

.026

)0.

026

(0.0

25)

0.01

8(0

.025

)0.

019

(0.0

28)

0.01

3(0

.028

)0.

018

(0.0

29)

0.01

4(0

.029

)0.

019

(0.0

27)

0.01

4(0

.028

)-0

.030

(0.0

30)

-0.0

39(0

.030

)In

itia

ltra

dera

tio

0.00

2(0

.026

)0.

004

(0.0

14)

0.01

4(0

.014

)0.

020

(0.0

14)

-0.0

04(0

.016

)0.

000

(0.0

17)

-0.0

09(0

.014

)-0

.005

(0.0

15)

0.00

2(0

.016

)0.

004

(0.0

17)

-0.0

02(0

.014

)0.

002

(0.0

14)

Bla

ckm

arke

tpr

emiu

m—

0.00

1(0

.002

)—

0.00

1(0

.002

)—

0.00

2(0

.003

)—

0.00

1(0

.002

)—

0.00

2(0

.003

)—

0.00

3(0

.002

)G

row

thra

teof

term

sof

trad

e

—0.

028

(0.0

52)

—0.

089*

(0.0

48)

—0.

064

(0.0

55)

—0.

085*

(0.0

51)

—0.

044

(0.0

56)

—-0

.010

(0.0

61)

R2

0.56

0.56

0.56

0.62

0.47

0.49

0.52

0.53

0.51

0.53

0.58

0.54

0.49

0.51

0.55

0.54

0.53

0.54

0.57

0.59

0.54

0.55

0.66

0.72

Obs

erva

tion

s29

2828

2728

2727

2627

2624

2329

2828

2728

2727

2627

2624

23R

egre

ssio

nte

st(p

valu

e)72

.009

(0.0

00)

72.2

20(0

.000

)84

.546

(0.0

00)

91.7

11(0

.000

)49

.005

(0.0

00)

50.3

04(0

.000

)57

.553

(0.0

00)

51.1

82(0

.000

)55

.232

(0.0

00)

56.7

317

(0.0

00)

73.0

73(0

.000

)75

.940

(0.0

00)

AR

(1)

(pva

lue)

-0.0

86(0

.643

)-0

.076

(0.6

88)

-0.1

55(0

.412

)-0

.154

(0.4

23)

0.11

5(0

.544

)0.

138

(0.4

72)

0.13

2(0

.494

)0.

177

(0.3

66)

0.06

4(0

.741

)0.

064

(0.7

45)

0.08

8(0

.667

)0.

088

(0.6

72)

Not

es:

See

Tab

le1.

As

the

coef

ficie

nts

for

Iva

riab

les

are

esse

ntia

llysi

mila

rto

wha

twe

foun

din

Tab

le1,

acco

rdin

gto

the

sugg

esti

onof

are

fere

ew

eha

veno

trep

orte

dth

emhe

refo

rbr

evit

y.



In comparing (4) with equations (1)–(3), it should be noted that tax revenueappeared as a conditioning, or I, variable in the earlier equations. However,as this is an element of the budget constraint, we include it in (4) as avariable in the set M. Consequently, there are now five rather than sixelements of I.

Equation (4) cannot be estimated due to the perfect collinearity betweenthe m elements Mj,it arising from the identity of the budget constraint. Con-sequently, (at least) one element Mj,it must be omitted. If, for simplicity, weassume Mm,it is the single omitted element, then the model to be estimatedbecomes

GR , ,it t jI

j it jM

j it itj

m

j

I M u= + + +=

−

=∑∑β β γ0

1

1

1

5

(5)

where, in relation to (4), γ β βj jM

mM= − . From standard results of linear

regression analysis, overall measures relating to the estimated regression(including R2, residuals etc.) and the coefficients β j

I are invariant to whichelement of the government budget constraint is excluded. However, the mag-nitude and significance of γ β βj j

MmM= − depends on both β j

M and βmM , and

therefore depends on which element is excluded. If, however, the excludedMm,it has coefficient βm

M = 0 , then γ βj jM= and the coefficient of each

included fiscal variable in (5) retains the same interpretation as in (4).Each of our models reported in Section 3 includes one government

expenditure category, together with tax revenue. Therefore, in attachingan estimated coefficient to a specific expenditure component, we implicitlyassumed all excluded β j

M = 0 . We now wish to acknowledge the possibilitythat the significant association between growth and each of the six com-ponents of public expenditure obtained in Section 3 could be the effect ofthe included expenditure component conditional on the fact that they arefinanced by the omitted ones. Indeed, by considering these one by one, anassociation of growth with one category could be spurious in the sense ofbeing attributable to other components of public expenditure with which it iscorrelated. To eliminate this possibility, we should ideally include all theelements of the government budget constraint, except for one category whosecoefficient we anticipate to be zero. Given our sample size, the scope forconducting such an exercise, however, is severely limited.

As a practical alternative, we consider the six components of publicexpenditure found to have significant impacts on growth in our earlier analysisin the context of three subgroups (total expenditure, total sectoral expendi-tures and sectoral investment expenditures). The elements of each subgroupare then included jointly in the model along with the budget constraint. Thus,by considering them jointly, we avoid possible spurious statistical significancearising due to correlation between included and excluded elements. Specifi-cally, we consider three models. In the first (model (6.1)), we include total



public current expenditure (CUR) and total public capital expenditure (CAP)along with the total expenditure net of outlays on total capital and currentexpenditure (OTHEXP(1)). In other words, OTHEXP(1) represents expendi-tures that are not classified as current or capital expenditure. In the secondregression (model (6.2)), we include variables IED and ITC to denote invest-ment expenditure in the education and in the transport and communicationsectors, respectively. Here, the variable OTHEXP(2) represents the totalexpenditure minus the outlays on investment expenditures in the educationand transport and communication sectors. Finally, in model (6.3) we includethe variables EDU, TC and DF to denote total expenditures in education,transport and communication, and defence sectors, respectively. As before,the variable OTHEXP(3) now represents total public expenditure net of theoutlay in the education, transport and communication, and defence sectors. Inall the models, expenditure variables are expressed as a percentage of GDP.

Specifically, the models are as follows:

GR CUR CAP OTHEXP,it t jI

j itM

itM

itM

itj

I= + + + + ( )

+=

∑β β β β β

β

0 1 2 31

5

4

1

MMit

Mit ituTX GD+ +β5 (6.1)

GR IED ITC OTHEXP,it t jI

j itM

itM

itM

itj

I= + + + + ( )

+=

∑β β γ γ γ

γ

0 1 2 31

5

4

2

MMit

Mit ituTX GD+ +γ 5 (6.2)

GR EDU TC DF

OTHEXP

,it t jI

j itM

itM

itM

itj

M

I= + + + +

+=

∑β β δ δ δ

δ

0 1 2 31

5

4 3(( ) + + +itM

itM

it ituδ δ5 6TX GD (6.3)

A few additional comments are necessary before we turn our attention tothe results. When considering models (6.1)–(6.3), we have seen that perfectcollinearity must be avoided by excluding an element of the budget con-straint. Ideally, one should omit a component which, according to the theory,has a neutral effect on growth. By including OTHEXP variables, we includethe expenditure side of the budget constraint, and we also explicitly includetax revenue (TX) and the budget surplus/deficit (GD), both as percentages ofGDP. Therefore, the element we choose to exclude from the models is non-tax revenue. This omission is based on the theoretical prediction (e.g. Barro,1990) that variation in non-distortionary revenue items is likely to generateinsignificant growth effects.15 Finally, our previous analysis indicates thatinclusion of the Z variables does not have any substantial impact on the

15It is worth noting that this assumption does not bias our results towards finding positive andsignificant growth effects from fiscal expenditures. In fact, if one assumes that non-taxrevenue is distortionary, our obtained coefficient would be an understatement of the actualeffect of the respective expenditure category on growth.



government expenditure coefficients. Consequently, we do not include thesevariables in models (6.1)–(6.3) on the ground of parsimony. Table 3 summa-rizes our results.

The effects of including the budget constraint, and also jointly consid-ering significant expenditure components, are strikingly evident from Table 3in comparison with Tables 1 and 2. In particular, of the six expenditurevariables, which were previously found to bear significant associations withgrowth, only three survive in the present analysis. These are total capitalexpenditure, total outlay in the education sector and investment expenditures

Table 3Growth Regressions with Budget Constraint

Total expenditureregression

Sectoralexpenditures

regression

Sectoralinvestmentsregression

M variablesCapital expenditure 0.151**

(0.059)— —

Current expenditure 0.093(0.057)

— —

Education expenditure — 1.582***(0.554)

0.658***(0.223)

Transport and communication expenditure — -0.001(0.237)

0.049(0.191)

Defence expenditure — — 0.021(0.111)

Other expenditures -0.059(0.719)

0.087**(0.040)

0.121**(0.054)

Tax revenue -0.006(0.054)

-0.009(0.059)

-0.209***(0.070)

Government surplus (/deficit) 0.146**(0.062)

0.153**(0.063)

0.156***(0.057)

I variablesPrivate investment 0.214***

(0.055)0.209***

(0.052)0.312***

(0.056)Initial GDP per capita 0.004

(0.003)0.006**

(0.003)0.010***

(0.003)Initial human capital -0.013**

(0.006)-0.011*(0.006)

-0.016***(0.005)

Initial life expectancy 0.089(0.067)

0.034(0.070)

0.055(0.063)

Political instability -0.016(0.019)

0.000(0.020)

-0.012(0.016)

R2 0.59 0.70 0.640.55 0.50 0.89

Observations 30 28 2130 28 21

Regression test (p value) 80.8(0.000)

105.1(0.000)

177.3(0.000)

AR(1) (p value) -0.019(0.918)

-0.129(0.495)

-0.092(0.673)

Notes: See Table 1.



in the education sector.16 Our results therefore point to education as the keysector for growth. In contrast, none of the variables related to defence and thetransport and communication sectors now show any significant associationwith growth. In the total expenditures model (6.1), we include current expen-diture to check whether this plays any role when considered in conjunctionwith capital expenditures, but it does not. It should also be noted that, in boththe total sectoral expenditures and sectoral investment expenditures models(the final two columns of Table 3), other expenditure has a significant andpositive coefficient (at a 5 per cent level of significance). However, given theset-up of our models, we cannot separately include all disaggregated expen-diture components for practical reasons, and therefore we cannot identifythose sectors that make this contribution.

The results on the growth effect of outlay on transport and communi-cation merit some additional comments. There is a general consensus amongempirical studies that the association between public investment expenditurein the transport and communication sector and growth is particularly strongand significant. For example, Aschauer (1989) finds that public investment inthe transport sector is highly correlated with private sector productivity in theUSA for the period 1949–85. Likewise, Easterly and Rebelo (1993) find thatpublic investment in this sector is consistently and positively correlated withgrowth. In Table 3, however, such evidence is absent.

We account for this on the basis of the two following observations. First,this difference may be due to the fact that, unlike previous studies, ouranalysis considers only developing countries. Second, and perhaps moresubstantively, this difference may reflect the presence of omission biases in theprevious studies due to their failure to consider the budget constraint andto consider more than one sector simultaneously.

In addition, our analysis brings out into the open the adverse growtheffects of government budget deficits.17 We find that these negative effects forthese countries are significant and of considerable magnitude: a 1 percentage

16It is interesting to note that some of our results are comparable with those found in theliterature that studied developed countries. In the case of sectoral expenditures for devel-oped countries, Miller and Russek (1997) find positive and significant coefficients (1.59 and1.86) for the education sector, while defence and transport and communication expendi-tures show no effect. For public investment (for either central government or total govern-ment), there are several studies that used developed country samples. For example, aftercontrolling for revenue component(s) of the government budget, Cashin (1995) finds thecoefficient for public investment to be significant with a low magnitude (i.e. 0.01) for allOrganization for Economic Cooperation and Development (OECD) countries, whileKocherlakota and Yi (1997) find it to be very high (0.14 or more) for the USA and UK.Although we follow the methodology of Kneller et al. (1999), we cannot directly comparetheir results with ours as their focus is different. That is, they consider expenditures to be ofproductive and unproductive nature and find productive expenditure with high positive andsignificant coefficients (0.26 or more). Finally, to the best of our knowledge, there is nostudy using a developed country sample to compare with our analysis for public invest-ments disaggregated by sectors.

17A similar view has been expressed by Fischer (1993).



point increase in the government surplus (as a percentage of GDP) is asso-ciated with an increase in the annual growth rate of real GDP per capita byan average of 0.15 percentage points. An increase in the budget deficit, ofcourse, has the corresponding negative effect.

We conclude this subsection with two additional observations. It isworth noting that the ‘Other expenditure’ coefficient appears insignificantin the first column in Table 3, whereas it appears significant in the next twocolumns. This result is not surprising as the ‘Other expenditure’ is notthe same variable across the three models. Further, the significance ofOTHEXP(2) and OTHEXP(3) in the last two columns indicates that thesevariables include some categories of expenditures that may be potentiallyimportant for growth. However, given our sample size, the scope for identi-fying these variables is limited. Finally, we note that in some cases thecoefficient of the same variable (e.g. tax revenue) assumes different values anddifferent levels of significance across the three models. This is due to differ-ences in the number of observations used across the models, resulting fromdata availability. When the three models are re-run using a common set ofcountries, this anomaly disappears while preserving the main conclusion ofour paper.18

4.2 Endogeneity Tests

In measuring the extent to which government expenditures affect economicgrowth, one has to recognize that fiscal and other economic variables evolvejointly: not only do government expenditures affect economic performance,but reverse causality is also a possibility. Therefore, we now turn to a verifi-cation of whether our results in Table 3 may be a manifestation of reversecausation or not. For this, we estimate the growth regression using three-stage least squares (3SLS).

In choosing the instruments for 3SLS, we follow the footsteps of Barroand Sala-i-Martin (1995, 1999). In particular, our set of instruments com-prises some of the original variables and lags of the other variables.19 In theabsence of data for government expenditure variables prior to 1970, we havechosen to run the regressions for the periods 1971–79 and 1981–89, instead of1970s–1980s, so that we obtain at least one set of observations for thegovernment expenditure variables that are predetermined for each equationof the system. Accordingly, the instruments for government expenditurevariables are their own observations for 1970 and 1980, respectively. Theselagged values are reasonable candidates for instruments since the correlation

18For this set of regressions we have used the same countries as used for model (6.3). Due tolimited space, we have not included the results of these regressions in the paper but they areavailable upon request.

19For comparison, we also considered agriculture’s share in GDP and population as instruments(e.g. Easterly and Rebelo, 1993). The results that we obtain are essentially similar to thosereported.



between the residuals in the growth regressions for two decades is small andinsignificant (Tables 1–3). Given that the initial variables (GDP per capita,human capital and life expectancy) are exogenous to the sample, these vari-ables enter as their own instruments. Finally, the instruments for privateinvestment and political instability are their averages for five years prior tothe specific decade. The results are reported in Table 4.20

For brevity, we report only the results for the coefficients of the sixgovernment expenditure variables that we considered previously. A straight-forward comparison of the results with those reported in Table 3 indicatesthat the sign of the coefficients and the levels of significance of the threesignificant expenditure variables (i.e. total capital expenditure, investment inthe education sector and the total outlay in the education sector) remainunaltered. Accordingly, to the extent that our instruments are valid, thegrowth effects of these three expenditure variables that we obtained in theprevious section should not be attributed to endogeneity. Further, as inTable 3, none of the other three components of expenditures is significant,with some of their coefficients being negative.

5 Conclusion

The objective of our study has been to evaluate the growth effects of publicexpenditures at the aggregate and disaggregate levels for 30 developing coun-tries. The primary contributions of this study are twofold. First, our innova-tion is to use more detailed data, and to analyse them while taking intoaccount the full implication of government budget constraints (as in Knelleret al., 1999). Second, our exclusive focus on developing countries is impor-

20The size of our sample has constrained our ability to include all components of total expendi-tures in model (6.3). In particular, we had to exclude the variable ‘Other expenditures’ fromthe model.

Table 4Endogeneity Test (3SLS)

Total expenditureregression

Sectoralexpenditures

regression

Sectoralinvestmentsregression

Capital expenditure 0.159**(0.063)

— —

Education expenditure — 2.245***(0.541)

1.093***(0.297)

Transport and communicationexpenditure

— -0.258(0.227)

0.013(0.206)

Defence expenditure — — -0.144(0.224)

Note: **Significant at 5 per cent; ***significant at 1 per cent.



tant, because the role of government expenditure for growth may profoundlydiffer across developed and developing countries.

Our analysis strongly suggests that government expenditure on educa-tion has long-lasting effects on economic prosperity. Underlying rationale forsuch a result is embedded not only in the new growth literature but also insome neoclassical growth models (e.g. Turnovsky, 2004) where transitionaleffects are very long lasting. Our result is robust with respect to whether weconsider total expenditure in education (in a regression that considers totalsectoral expenditures) or investment in education (in a model that focuses onsectoral investment expenditures). Such strong evidence is absent in the exist-ing empirical literature. Accordingly, from the policy perspective, our analy-sis prioritizes the allocation of scarce government resources towards theeducation sector for at least some developing countries. Further, our analysisalso suggests that aggregate current expenditure has no effect on growth,whereas aggregate capital expenditure has a positive effect. This implies that,for developing countries, decisions on current versus capital expenditureshould (at least in the aggregate) favour the latter in order to enhance growth.

Our results should not, however, be interpreted as implying that expen-diture on education or on capital projects should be increased irrespective ofhow these are financed. Indeed, our analysis is careful in considering the roleof the government budget constraint. Since tax revenue has a negative impact(although not always significant) on growth, while increasing the governmentdeficit has a highly significant negative effect, the raising of additional financewill moderate the positive effects of education or capital expenditure. Perhapsthe importance of our results can be considered most clearly in the context ofa transfer of, say, 1 percentage point of government expenditure in relation toGDP from another sector towards education, or from current to capitalexpenditure, where our results imply that such a transfer will be growthenhancing.

Appendix

The Data

Countries Included (According to the World Bank Classification of Regions). Centraland North America (1), Bahamas; South Asia (5), Bangladesh, India, Nepal,Pakistan, Sri Lanka; Sub-Saharan Africa (16), Botswana, Burundi, Congo, Ethiopia,Ghana, Kenya, Madagascar, Malawi, Mauritius, Nigeria, Rwanda, Sierra Leone,Sudan, Tanzania, Zaire, Zambia; Middle East and North Africa (3), Morocco, Syria,Tunisia; Latin America and the Caribbean (2), Guatemala, Jamaica; East Asia andthe Pacific (3), Indonesia, Malaysia, Thailand.

Definitions and Sources. Definitions for all variables and data sources are presentedin Table A1.



Table A1Definition of Variables and Data Sources

Variable Data sources

GDP datagr Average growth rate in GDP per capita World Bank CDROMlgc Log of GDP per capita World Bank CDROM

Government expenditure categoriescur Government current expenditure (% of GDP) World Bank Reports and IMFcap Government capital expenditure (% of GDP) World Bank Reports and IMFcdf Government consumption in defence (% of GDP) World Bank Reports and IMFced Government consumption education (% of GDP) World Bank Reports and IMFchl Government consumption in health (% of GDP) World Bank Reports and IMFcag Government consumption in agriculture

(% of GDP)World Bank Reports and IMF

cmf Government consumption in manufacturing(% of GDP)

World Bank Reports and IMF

ctc Government consumption in transport andcommunication (% of GDP)


idf Government investment in defence (% of GDP) World Bank Reports and IMFied Government investment in education (% of GDP) World Bank Reports and IMFihl Government investment in health (% of GDP) World Bank Reports and IMFiag Government investment in agriculture (% of GDP) World Bank Reports and IMFimf Government investment in manufacturing (% of

GDP)World Bank Reports and IMF

itc Government investment in transport andcommunication (% of GDP)


df Government expenditure in defence (% of GDP) World Bank Reports and IMFedu Government expenditure education (% of GDP) World Bank Reports and IMFhl Government expenditure in health (% of GDP) World Bank Reports and IMFag Government expenditure in agriculture (% of GDP) World Bank Reports and IMFmf Government expenditure in manufacturing

(% of GDP)World Bank Reports and IMF

tc Government expenditure in transport andcommunication (% of GDP)


Other variablesp Primary school enrolment ratio Barro and Lee (1994)s Secondary school enrolment ratio Barro and Lee (1994)h Higher education enrolment ratio Barro and Lee (1994)psh A linear combination of p, s and h (see below) Barro and Lee (1994)life Log of life expectancy Barro and Lee (1994)as No. of assassinations per million population per

yearBarro and Lee (1994)

rev No. of revolutions per year Barro and Lee (1994)coup No. of coups per year Barro and Lee (1994)pinst A linear combination of as, rev and coup

(see below)Barro and Lee (1994)

bmp Black market premium Barro and Lee (1994)m2 Broad money (M2) (% of GDP) World Bank CDROMtr Trade ratio (export plus import as % of GDP) Barro and Lee (1994)tt Growth rate of terms of trade Barro and Lee (1994)tx Tax revenue (% of GDP) Government Finance Statistics,

IMFgsd Government surplus/deficit (% of GDP) World Bank CDROMpviw Private investment (% of GDP) Barro and Lee (1994)agr Agriculture’s value added (% of GDP) World Bank CDROMpop Log of population Barro and Lee (1994)



Human capital (PSH):Following Landau (1983), we construct the initial human capital (PSH) variable as theweighted sum of the initial enrolment ratios (per cent) in primary and secondaryschools and in higher education. The weights are 1 for primary school enrolment ratio,2 for secondary school and 3 for enrolment in higher education. The weights areapproximations to the relative values of three types of education. The PSH variable isnecessary because of the high multicollinearity between the separate enrolment rates.The data for average schooling years are missing for one-fourth of the countries in thesample; thus the enrolment rates are probably better available measures of investmentin education. The other rationale for taking enrolment rates is that these are morefrequently used in the literature (see Easterly and Rebelo, 1993; Barro and Sala-i-Martin, 1995, 1999; among others).

Political instability (PINST):Following Barro and Sala-i-Martin (1995, 1999), we take the average of each decadeof revolutions and coups per year and political assassinations per million inhabitantsper year.

Summary Statistics. Table A2 presents summary statistics for the variables usedin the results reported in the paper. Data are used primarily as decade averages,relating to the 1970s and 1980s. However, for 3SLS, we take the averages for 1971–80and 1981–90 instead of 1970–80 and 1980–90, respectively. A suffix of two numbersafter a variable name indicates a specific year (e.g. P70 is the primary school enrolmentratio in 1970), while a single number refers to the period for a specific average; forexample, gr1 is the average growth rate of GDP per capita for 1970–80, gr2 is for1980–90, gr3 is for 1971–80 and gr4 is for 1981–90.

Table A2Summary Statistics

Variable Observations Mean Std Dev. Min Max

gr1 30 0.024 0.028 -0.022 0.111gr2 30 0.012 0.023 -0.023 0.071lgc70 30 3.378 0.816 2.087 5.295lgc80 30 3.490 0.818 2.222 5.518psh70 30 1.082 0.598 0.246 2.821psh80 30 1.501 0.650 0.368 3.248life70 30 1.694 0.068 1.535 1.827life80 30 1.729 0.067 1.581 1.853pinst1 30 0.097 0.116 0 0.445pinst2 30 0.097 0.154 0 0.730bmp1 29 0.421 0.473 0 2.024bmp2 30 0.837 1.423 0 7.185m270 30 0.239 0.107 0.085 0.435m280 30 0.320 0.142 0.079 0.775tr70 29 0.445 0.242 0.077 0.925tr80 30 0.600 0.333 0.157 1.333tt1 28 0.014 0.070 -0.085 0.176tt2 28 -0.021 0.027 -0.106 0.011tx1 30 0.144 0.050 0.045 0.257



Table A2Continued

Variable Observations Mean Std Dev. Min Max

tx2 30 0.150 0.062 0.056 0.284gd1 30 -0.052 0.033 -0.139 0.002gd2 30 -0.056 0.044 -0.132 0.114pvi1 30 0.119 0.058 0.026 0.316pvi2 30 0.109 0.049 0.034 0.217cur1 30 0.140 0.060 0.012 0.251cur2 30 0.159 0.069 0.011 0.304cdf1 29 0.025 0.026 0.001 0.136cdf2 28 0.028 0.030 0.002 0.142ced1 30 0.025 0.015 0.002 0.056ced2 29 0.026 0.016 0.002 0.061chl1 30 0.010 0.007 0.001 0.026chl2 29 0.010 0.008 0.001 0.032cag1 30 0.008 0.006 0.001 0.029cag2 30 0.007 0.005 0.001 0.021cmf1 23 0.002 0.002 0 0.008cmf2 23 0.003 0.007 0 0.033ctc1 28 0.007 0.005 0.001 0.022ctc2 28 0.004 0.003 0 0.011cap1 30 0.072 0.038 0.004 0.160cap2 30 0.086 0.049 0.006 0.172idf1 28 0.003 0.005 0 0.018idf2 25 0.002 0.005 0 0.018ied1 30 0.006 0.005 0.000 0.019ied2 29 0.007 0.006 0.001 0.021ihl1 30 0.002 0.002 0.000 0.009ihl2 29 0.004 0.005 0.000 0.027iag1 30 0.011 0.007 0.001 0.028iag2 30 0.014 0.012 0.001 0.050imf1 28 0.008 0.013 0.000 0.066imf2 28 0.010 0.012 0 0.047itc1 29 0.016 0.012 0.001 0.046itc2 29 0.017 0.014 0.001 0.066te1 30 0.212 0.087 0.017 0.390te2 30 0.246 0.103 0.019 0.443df1 28 0.027 0.024 0.000 0.116df2 25 0.026 0.022 0.002 0.100edu1 30 0.032 0.015 0.004 0.058edu2 28 0.033 0.016 0.004 0.067hl1 30 0.012 0.007 0.001 0.027hl2 28 0.014 0.008 0.001 0.035ag1 30 0.019 0.011 0.002 0.049ag2 30 0.021 0.014 0.002 0.059mf1 23 0.010 0.014 0.000 0.067mf2 23 0.014 0.014 0 0.050tc1 28 0.023 0.015 0.002 0.069tc2 28 0.022 0.016 0.002 0.074agr70 29 0.331 0.163 0.07 0.669agr80 29 0.292 0.144 0.08 0.579pop70 29 4.005 0.655 2.79 5.744pop80 29 4.119 0.656 2.95 5.838popgr70 30 2.572 0.586 1.31 3.609popgr80 30 2.575 0.658 0.98 3.577



References

Alexander, W. R. (1990). ‘Growth: Some Combined Cross-sectional and Time-seriesEvidence from OECD Countries’, Applied Economics, Vol. 22, pp. 1197–1204.

Aschauer, D. (1989). ‘Is Government Spending Productive?’, Journal of MonetaryEconomics, Vol. 23, pp. 177–200.

Azariadis, C. and Drazen, A. (1990). ‘Threshold Externalities in Economic Develop-ment’, Quarterly Journal of Economics, Vol. 105, No. 2, pp. 501–526.

Barro, R. J. (1990). ‘Government Spending in a Simple Model of EndogenousGrowth’, Journal of Political Economy, Vol. 98, No. 5, pp. S103–S125.

Barro, R. J. (1991). ‘Economic Growth in a Cross Section of Countries’, QuarterlyJournal of Economics, Vol. CVI, pp. 407–444.

Barro, R. J. and Lee, J.-W. (1994). ‘Sources of Economic Growth’, Carnegie-Rochester Conference Series on Public Policy, Vol. 40, pp. 1–46.

Barro, R. J. and Sala-i-Martin, X. (1995). Economic Growth, Boston, MA, McGraw-Hill.

Barro, R. J. and Sala-i-Martin, X. (1999). Economic Growth, 1st edn, Cambridge, MA,MIT Press.

Benhabib, J. and Spiegel, M. M. (1994). ‘The Role of Human Capital in EconomicDevelopment: Evidence from Aggregate Cross-country Data’, Journal of Mon-etary Economics, Vol. 34, No. 2, pp. 143–173.

Benoit, E. (1978). ‘Growth and Defense in Developing Countries’, Economic Devel-opment and Cultural Change, Vol. 26, pp. 271–280.

Biswas, B. and Ram, R. (1986). ‘Military Expenditures and Economic Growth in LessDeveloped Countries: an Augmented Model and Further Evidence’, EconomicDevelopment and Cultural Change, Vol. 34, pp. 362–372.

Cashin, P. (1995). ‘Government Spending, Taxes, and Economic Growth’, IMF StaffPapers, Vol. 42, No. 2, pp. 237–269.

Deger, S. and Smith, R. (1983). ‘Military Expenditure and Growth in Less DevelopedCountries’, Journal of Conflict Resolution, Vol. 27, pp. 335–353.

DeLong, J. B. and Summers, L. H. (1991). ‘Equipment Investment and EconomicGrowth’, Quarterly Journal of Economics, Vol. 106, No. 2, pp. 445–502.

Devarajan, S., Swaroop, V. and Zou, H. (1996). ‘The Composition of Public Expen-diture and Economic Growth’, Journal of Monetary Economics, Vol. 37,pp. 313–344.

Easterly, W. and Rebelo, S. (1993). ‘Fiscal Policy and Economic Growth: an Empiri-cal Investigation’, Journal of Monetary Economics, Vol. 32, pp. 417–458.

Fischer, S. (1993). ‘The Role of Macroeconomic Factors in Growth’, Journal ofMonetary Economics, Vol. 32, pp. 485–512.

Fredriksen, P. and Looney, R. (1982). ‘Defense Expenditures and Economic Growthin Developing Countries: Some Further Empirical Evidence’, Journal of Eco-nomic Development, Vol. 7, pp. 113–125.

de la Fuente, A. (1997) ‘Fiscal Policy and Growth in OECD’, CEPR Discussion Paper1755.

Gemmell, N. (1996). ‘Evaluating the Impacts of Human Capital Stocks and Accumu-lation on Economic Growth’, Oxford Bulletin of Economics and Statistics, Vol. 1,No. 58, pp. 9–28.

Grier, K. B. and Tullock, G. (1989). ‘An Empirical Analysis of Cross-national Eco-nomic Growth, 1951–1980’, Journal of Monetary Economics, Vol. 24, pp. 259–276.

Helms, L. (1985). ‘The Effect of State and Local Taxes on Economic Growth: a TimeSeries Cross-section Approach’, Review of Economics and Statistics, Vol. 67,No. 3, pp. 574–582.



Hulten, C. (1996). ‘Infrastructure Capital and Economic Growth: How Well You UseIt May Be More Important than How Much You Have’, NBER Working Paper5847.

IMF (1986). A Manual on Government Finance Statistics, Washington, DC, IMF.King, R. G. and Levine, R. (1993). ‘Finance, Entrepreneurship and Growth:

Theory and Evidence’, Journal of Monetary Economics, Vol. 32, pp. 513–542.

Kneller, R. Bleaney, M. F. and Gemmell, N. (1999). ‘Fiscal Policy and Growth:Evidence from OECD Countries’, Journal of Public Economics, Vol. 74,pp. 171–190.

Knight, M., Loayza, N. and Villanueva, D. (1996). ‘The Peace Dividend:Military Spending Cuts and Economic Growth’, World Bank Policy ResearchWorking Paper 1577, February.

Kocherlakota, N. and Yi, K. (1997). ‘Is there Endogenous Long-run Growth? Evi-dence from US and UK’, Journal of Money, Credit and Banking, Vol. 29, No. 2,pp. 235–262.

Kormendi, R. C. and Meguire, P. G. (1985). ‘Macroeconomic Determinantsof Growth: Cross-country Study’, Journal of Monetary Economics, Vol. 16,pp. 141–163.

Landau, D. L. (1983). ‘Government Expenditure and Economic Growth: a CrossCountry Study’, Southern Economic Journal, Vol. 49, pp. 783–792.

Landau, D. L. (1985). ‘Government Expenditure and Economic Growth in the Devel-oped Countries, 1952–76’, Public Choice, No. 3, pp. 459–477.

Landau, D. L. (1986). ‘Government and Economic Growth in the Less DevelopedCountries: an Empirical Study for 1960–88’, Economic Development and CulturalChange, Vol. 35, pp. 35–75.

Levine, R. and Renelt, D. (1992). ‘A Sensitivity Analysis of Cross-countryGrowth Regressions’, American Economic Review, Vol. 82, No. 4, pp. 942–963.

Mankiw, N. G., Romer, D. and Weil, D. N. (1992). ‘A Contribution to the Empiricsof Economic Growth’, Quarterly Journal of Economics, Vol. 107, No. 2,pp. 407–437.

Miller, S. M. and Russek, F. S. (1997). ‘Fiscal Structures and Economic Growth’,Economic Inquiry, Vol. 35, pp. 603–613.

Mofidi, A. and Stone, J. (1990). ‘Do State and Local Taxes affect Economic Growth?’,Review of Economics and Statistics, Vol. 72, No. 4, pp. 686–691.

Pritchett, L. (2001). ‘Where Has All the Education Gone?’, World Bank EconomicReview, Vol. 15, pp. 367–391.

Ram, R. (1986). ‘Government Size and Economic Growth: a New Framework andSome Evidence from Cross-section and Time Series Data’, American EconomicReview, Vol. 76, pp. 191–203.

Rodrik, D. (1998). ‘Why Do More Open Economies Have Bigger Governments?’,Journal of Political Economy, Vol. 106, No. 5, pp. 997–1034.

Romer, P. (1989). ‘What Determines the Rate of Growth and Technological Change’,World Bank Working Papers.

Romer, P. M. (1990). ‘Human Capital and Growth: Theory andEvidence’, Carnegie-Rochester Conference Series on Public Policy, Vol. 40,pp. 47–57.

Temple, J. (1999a). ‘The New Growth Evidence’, Journal of Economic Literature,Vol. 37, pp. 112–156.

Temple, J. (1999b). ‘A Positive Effect of Human Capital on Growth’, EconomicsLetters, Vol. 65, pp. 131–134.



Turnovsky, S. J. (2004). ‘The Transitional Dynamics of Fiscal Policy: Long-runCapital Accumulation and Growth’, Journal of Money, Credit and Banking,Vol. 36, No. 5, pp. 883–910.

World Bank (1988). World Development Report, World Bank, Oxford UniversityPress.



public expenditure and economic growth: a disaggregated analysis for developing countries*

Documents

growth theory

economic growth

government size

disaggregated level

omitted variables

government budget constraint

elements ofpublic expenditure

blackwell publishing